Coming from pandas

pandasx follows pandas closely: same concepts, same semantics for missing data and alignment, same function names in camelCase. The differences come from JavaScript itself — it has no operator overloading and no obj[key] = value hooks — so those forms become methods.

Naming

Every snake_case name becomes camelCase, and keyword arguments become an options object:

pandas	pandasx
pd.read_csv('a.csv', index_col='id')	pd.readCsv('a.csv', { indexCol: 'id' })
df.sort_values('a', ascending=False)	df.sortValues('a', { ascending: false })
df.drop_duplicates(subset=['a'])	df.dropDuplicates({ subset: ['a'] })
df.groupby('k', as_index=False)	df.groupby('k', { asIndex: false })
s.value_counts(normalize=True)	s.valueCounts({ normalize: true })

Indexing with []

pandas	pandasx
df['a']	df.get('a')
df[['a', 'b']]	df.get(['a', 'b'])
df[df.a > 1]	df.get(df.get('a').gt(1)) or df.loc(df.get('a').gt(1))
df['c'] = values	df.set('c', values)
df.loc['x', 'a']	df.loc('x', 'a')
df.loc['x':'z']	df.loc(pd.slice('x', 'z')) (inclusive, like pandas)
df.iloc[1:3, 0]	df.iloc(pd.slice(1, 3), 0)
df.loc[:, 'a':'c']	df.loc(null, pd.slice('a', 'c'))
df.loc[mask, 'a'] = 0	df.setLoc(mask, 'a', 0)
df.at['x', 'a'] = 1	df.setAt('x', 'a', 1)
df[('price', 'min')]	df.column(['price', 'min'])

Operators

pandas	pandasx
s + 1, s - t, s * 2, s / t	s.add(1), s.sub(t), s.mul(2), s.div(t)
s // 2, s % 2, s ** 2	s.floordiv(2), s.mod(2), s.pow(2)
s == 1, s != 1, s < 1, s >= 1	s.eq(1), s.ne(1), s.lt(1), s.ge(1)
a & b, a | b, ~a, a ^ b	a.and(b), a.or(b), a.not(), a.xor(b)
-s	s.neg()

Like pandas, binary operations align on the index:

import * as pd from 'pandasx';

const a = new pd.Series([1, 2, 3], { index: ['x', 'y', 'z'] });
const b = new pd.Series([10, 20], { index: ['y', 'z'] });
console.log(a.add(b));

x   NaN
y  12.0
z  23.0
dtype: float64

String expressions still work

query and eval accept the same strings as pandas:

import * as pd from 'pandasx';

const df = new pd.DataFrame({ a: [1, 5, 10], b: ['x', 'y', 'x'] });
const limit = 3;
console.log(df.query('a > @limit and b == "x"', { locals: { limit } }));
console.log(df.eval('c = a * 2'));

    a  b
2  10  x
    a  b   c
0   1  x   2
1   5  y  10
2  10  x  20

Row-wise functions

df.apply(fn, { axis: 1 }) passes each row as a plain object, which is more natural in JavaScript than a Series:

import * as pd from 'pandasx';

const df = new pd.DataFrame({ w: [2, 3], h: [5, 4] });
console.log(df.apply((row) => (row.w as number) * (row.h as number), { axis: 1 }));

0  10
1  12
dtype: int64

Values and missing data

• Missing numbers are NaN; missing strings, dates and objects read back as null.
• The nullable dtypes Int64, Float64 and boolean read missing values as null (pandas' pd.NA); pd.NA is exported as null.
• Datetimes are JavaScript Date objects; durations are pd.Timedelta, periods are pd.Period.
• Integers beyond 2⁵³ use the bigint dtype.

Asynchronous I/O

CSV, JSON, HTML and XML readers are synchronous like pandas. Readers that depend on asynchronous libraries — Excel, Parquet, Feather, HDF5 and SQL — return a Promise:

const df = await pd.readParquet('data.parquet');
await df.toExcel('report.xlsx');

See also Differences from pandas for the few behaviours that deliberately differ.